Differentiable plasticity: training plastic neural networks with backpropagation

TL;DR

This paper demonstrates that large recurrent neural networks with differentiable plasticity, optimized via backpropagation, can effectively learn, memorize, and adapt in complex tasks, outperforming non-plastic networks in various learning scenarios.

Contribution

It introduces a method to train large recurrent networks with plastic connections using gradient descent, enabling efficient lifelong learning and meta-learning capabilities.

Findings

Plastic networks can memorize high-dimensional images not seen during training.

They perform competitively on meta-learning tasks like Omniglot.

In reinforcement learning, plastic networks outperform non-plastic counterparts.

Abstract

How can we build agents that keep learning from experience, quickly and efficiently, after their initial training? Here we take inspiration from the main mechanism of learning in biological brains: synaptic plasticity, carefully tuned by evolution to produce efficient lifelong learning. We show that plasticity, just like connection weights, can be optimized by gradient descent in large (millions of parameters) recurrent networks with Hebbian plastic connections. First, recurrent plastic networks with more than two million parameters can be trained to memorize and reconstruct sets of novel, high-dimensional 1000+ pixels natural images not seen during training. Crucially, traditional non-plastic recurrent networks fail to solve this task. Furthermore, trained plastic networks can also solve generic meta-learning tasks such as the Omniglot task, with competitive results and little parameter overhead. Finally, in reinforcement learning settings, plastic networks outperform a non-plastic equivalent in a maze exploration task. We conclude that differentiable plasticity may provide a powerful novel approach to the learning-to-learn problem.